Blind separation of two multi-level sources from a single linear mixture

We present a novel method for the blind separation of two multi-level sources from one observed instantaneous linear mixture. Our approach is based on the geometric characteristics of the cluster of points and in particular in the distribution of the distances between the cluster centers. In the noiseless case the core of the method is conceptually very simple, deterministic, non-iterative and fast. In the noisy case, the core-algorithm must be preceded by a clustering scheme, for example, the EM algorithm in order to identify the centers. The method works directly on the time domain and so no signal transformation is required.